Dropbox Block-Level Sync Architecture

Dropbox splits files into blocks (typically 4MB) and computes a SHA-256 hash of each block. The hash serves as both the block's identifier and its integrity check. Two identical blocks — even across different files or different users — produce the same hash and are stored only once. This content-addressable storage provides massive deduplication: common files (OS installers, popular PDFs) are stored once globally but appear in millions of accounts. When uploading, the client first sends block hashes to the server; if a hash already exists, the block doesn't need to be uploaded at all.

When a user modifies a file, the desktop client re-chunks the file and computes new block hashes. Only blocks whose hashes have changed are uploaded — typically a small fraction of the total file. For a 2GB presentation where one slide was edited, perhaps only one or two 4MB blocks need transmission. The Metadata Server is updated with the new block list for that file version. Dropbox also applies streaming compression (LZ4) to blocks before transmission, further reducing bandwidth. For tiny edits within a block, sub-block binary diffing can reduce the transfer to just the changed bytes.

The Metadata Server maintains the authoritative mapping of file paths to block lists, along with version history, permissions, and sharing metadata. It's backed by a sharded MySQL cluster with write-ahead logging for durability. Every file operation (create, modify, delete, move, rename) generates a journal entry with a monotonically increasing server-side sequence number (cursor). Clients poll for journal entries since their last known cursor to discover changes. This cursor-based sync protocol ensures clients never miss an update and can resume sync after any disconnection by replaying from their last cursor.

When two devices edit the same file before syncing, a conflict occurs. Dropbox uses a last-writer-wins strategy based on server-side timestamps — the later modification becomes the canonical version. However, the 'losing' version is not silently discarded: Dropbox creates a 'conflicted copy' file (named with the device and timestamp) so users can manually reconcile differences. For collaborative editing scenarios, Dropbox Sync APIs provide operational transform support. This approach prioritizes data preservation over automatic resolution — no user data is ever silently lost.

If two Dropbox clients are on the same local network, LAN Sync enables direct peer-to-peer block transfer without routing through Dropbox's servers. Clients broadcast their presence on the local network via UDP, discover peers with shared folders, and transfer blocks directly over the LAN at full local network speed (typically 100Mbps–1Gbps vs. potentially slow internet upload). This is transformative in office environments where multiple team members share the same folders — large file syncs complete in seconds instead of minutes. The feature reduces Dropbox's bandwidth costs while dramatically improving user experience.